Microsoft Tests AI with Fake Companies to Keep Your Data Private

Microsoft wants to make sure its AI assistant, Copilot, works well for you without ever looking at your private emails or documents. So, they built fake companies—complete with made-up employees, meetings, and projects—to test Copilot. These fake companies are called 'synthetic environments' and are designed to act like real organizations. This way, Microsoft can see how Copilot handles realistic tasks, like summarizing a meeting or finding a document, without ever touching your actual data.

The fake companies are not just random; they're based on real-world patterns from public data. They even change over time, just like real companies, with new emails and meetings appearing. This is important because many questions you ask Copilot are about time—like 'What did we decide last week?'—and the fake data needs to reflect that. Microsoft also created versions for different industries, like healthcare, so Copilot can be tested on the specific questions people in those fields ask.

To test Copilot thoroughly, Microsoft needs lots of realistic questions. Writing these by hand would be slow and expensive, so they built an automated system that turns simple ideas into detailed, industry-specific questions. This system uses a six-step process to generate thousands of queries that mimic what real users might ask. The result is a rigorous evaluation that helps Microsoft improve Copilot's accuracy and usefulness.

For you, this means Copilot can get better at helping you without ever compromising your privacy. Your data stays yours, and Microsoft can still ensure Copilot is reliable. So next time Copilot summarizes your meeting or drafts an email, you can trust it's been tested on fake data that looks just like your real work—keeping your information safe.